This talk delved into all the aspects and challenges associated with training large language models (LLMs). The talk covered everything from gathering data at scale, optimizing training performance on large GPU clusters, to running meaningful evaluations. In addition to the technical aspects participants also got an overview of the governance of such an endeavour and how such models can be built and released to benefit all of society.

You can download Leandro's presentation slides and watch a video of his talk below: